Leading encoding comments creeps into RMarkdown file

[alcrene]

This is a somewhat opposite problem to #241. When the Jupytext metadata includes the file encoding, the leading comment for Python scripts is erroneously included in RMarkdown files.

Steps to reproduce in JupyterLab:

1. Create a notebook. A single markdown cell, with say "α", will suffice. The relevant Jupytext metadata is

 "metadata": {
  "jupytext": {
   "formats": "ipynb,Rmd"
  },

2. Pair the notebook with an RMarkdown file. Note that the export is fine; the RMarkdown looks as follows:

---
jupyter:
  jupytext:
    formats: ipynb,Rmd
    text_representation:
      extension: .Rmd
      format_name: rmarkdown
      format_version: '1.2'
      jupytext_version: 1.5.0
---

α

3. Add an encoding entry to the metadata. One way to do this is:
   1. Additionally pair the notebook with a Python script. It will include the line # -*- coding: utf-8 -*- because of the unicode character.
   2. Edit the Python script (e.g. write "αb") and save.
   3. Reopen the notebook in Jupyter and save. A dialog will ask to overwrite; click yes. Now the notebook metadata reads

 "metadata": {
  "jupytext": {
   "encoding": "# -*- coding: utf-8 -*-",
   "formats": "ipynb,Rmd,py:light"
  },

4. In addition, the RMarkdown is now invalid because it begins with the Python encoding line:

# -*- coding: utf-8 -*-
---
jupyter:
  jupytext:
    formats: ipynb,Rmd,py:light
    text_representation:
      extension: .Rmd
      format_name: rmarkdown
      format_version: '1.2'
      jupytext_version: 1.5.0
---

αb

[mwouts]

Thank you @alcrene for the detailed report! That is very helpful. I think I should be able to reproduce this unwanted behavior easily in a test (given that you've done all the work above 👍). I will let you know when the fix is available.

[mwouts]

Hi @alcrene , this should be solved in the development version. If you wish, you can give it a try with

pip install git+https://github.com/mwouts/jupytext.git

Thank you again for reporting it!

[alcrene]

Thank you for the rapid fix ! I just tested with my original use case, and as far as I can tell it works as intended.